![[2508.19113] Hybrid Deep Searcher: Scalable Parallel and Sequential Search Reasoning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2508.19113] Hybrid Deep Searcher: Scalable Parallel and Sequential Search Reasoning
Summary
The paper presents HybridDeepSearcher, a novel approach that enhances search reasoning by integrating parallel query expansion with structured evidence aggregation, significantly improving performance on various benchmarks.
Why It Matters
As AI continues to evolve, the ability to perform scalable reasoning with external knowledge is crucial. This research addresses limitations in existing models, offering a more effective solution for deep search tasks, which is essential for advancements in AI applications.
Key Takeaways
- HybridDeepSearcher combines parallel query expansion with structured aggregation for improved reasoning.
- The new approach outperforms state-of-the-art methods, showing significant F1 score improvements.
- Test-time search scaling is consistent, enhancing performance with additional search turns.
Computer Science > Artificial Intelligence arXiv:2508.19113 (cs) [Submitted on 26 Aug 2025 (v1), last revised 24 Feb 2026 (this version, v2)] Title:Hybrid Deep Searcher: Scalable Parallel and Sequential Search Reasoning Authors:Dayoon Ko, Jihyuk Kim, Haeju Park, Sohyeon Kim, Dahyun Lee, Yongrae Jo, Gunhee Kim, Moontae Lee, Kyungjae Lee View a PDF of the paper titled Hybrid Deep Searcher: Scalable Parallel and Sequential Search Reasoning, by Dayoon Ko and 8 other authors View PDF Abstract:Large reasoning models (LRMs) combined with retrieval-augmented generation (RAG) have enabled deep research agents capable of multi-step reasoning with external knowledge retrieval. However, we find that existing approaches rarely demonstrate test-time search scaling. Methods that extend reasoning through single-query sequential search suffer from limited evidence coverage, while approaches that generate multiple independent queries per step often lack structured aggregation, hindering deeper sequential reasoning. We propose a hybrid search strategy to address these limitations. We introduce HybridDeepSearcher, a structured search agent that integrates parallel query expansion with explicit evidence aggregation before advancing to deeper sequential reasoning. To supervise this behavior, we introduce HDS-QA, a novel dataset that guides models to combine broad parallel search with structured aggregation through supervised reasoning-query0retrieval trajectories containing parallel sub-queries...
